Top 3 Under-Hyped AI Capabilities

The recent AWS Summit in Sydney made one thing clear: GenAI now dominates the tech landscape at every level. Just a few years ago, GenAI-related sessions were rare; today, you're hard pressed to find a session that doesn't mention it. Experiencing the state of the products first-hand, from GenAI Golf to GenAI Portfolio Optimisation, and speaking to the devs about “what’s under the hood”, let me tell you what's often not spoken about, and under-hyped about the most powerful products today:

1. Predictive AI Models

These are not GenAI. They’re well-established robust machine learning methods, backed by big numerical datasets to predict real-world outcomes. Examples being risk prediction, pricing prediction, and behaviour prediction. During the Keynote speech, Anton de Weger spoke about how his AI Health startup has reduced the time between diagnosis and treatment for epilepsy by 15 years using predictive AI.

2. Text-to-speech (TTS) models

In parallel to the LLMs, TTS models have undergone a huge leap in capabilities to unlock LLM’s audio capabilities. Whenever you listen to your Claude or ChatGPT speak, there’s a secondary TTS model involved. Elevenlabs and OpenAI are leading the way in this technology, but there are open-source models that don’t trail too far behind. With five seconds of audio, you can make a near perfect replica of your own voice.

3. Diffusion Models

Diffusion models are the backbone of image and video generation capabilities. (The above image was created by AI)

Similar to TTS models they are undergoing tremendous improvement. When you’re asking ChatGPT to make you an image or video, there’s an LLM creating a text prompt then sending it to a diffusion model to create frame(s) accordingly. Currently, Google is cooking up a “Gemini Diffusion” LLM. This is an attempt to apply a new method to text generation, which could change the game completely if successful.

In conclusion, think twice before getting caught up in the hype around futuristic "human-like" agents.

Instead, consider the powerful AI capabilities that exist today.